Content validity index (CVI): I-CVI and S-CVI calculation
The content validity index (CVI) is the proportion of experts who rate an item as relevant. Experts rate each item on a 4-point relevance scale (1 = not relevant, 2 = somewhat relevant, 3 = quite relevant, 4 = highly relevant). Ratings of 3 or 4 count as agreement.
- I-CVI (item level) = number of experts rating the item 3 or 4 ÷ number of experts.
- S-CVI/Ave (scale level, average) = the mean of the I-CVIs across items.
- S-CVI/UA (universal agreement) = proportion of items rated 3 or 4 by every expert.
Commonly cited standards: I-CVI of 1.00 with five or fewer experts and at least .78 with more (Lynn 1986; Polit & Beck 2006), S-CVI/Ave of .90 or higher (Polit & Beck 2006). Polit, Beck & Owen (2007) also recommend reporting a modified kappa to adjust I-CVI for chance agreement.
Formulas
The modified kappa (Polit, Beck & Owen 2007) is interpreted with the criteria they adopted from Cicchetti & Sparrow (1981): fair .40 to .59, good .60 to .74, excellent above .74.
Worked example: six experts, five items
Illustrative ratings, not real data. E1 to E6 are experts.
| Item | E1 | E2 | E3 | E4 | E5 | E6 | A (rated 3-4) | I-CVI | Pc | k* | Rating |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Item 1 | 4 | 4 | 3 | 4 | 4 | 3 | 6 | 1.00 | 0.016 | 1.00 | Excellent |
| Item 2 | 4 | 3 | 4 | 2 | 4 | 4 | 5 | 0.83 | 0.094 | 0.82 | Excellent |
| Item 3 | 3 | 2 | 4 | 4 | 1 | 3 | 4 | 0.67 | 0.234 | 0.56 | Fair |
| Item 4 | 4 | 4 | 4 | 4 | 3 | 4 | 6 | 1.00 | 0.016 | 1.00 | Excellent |
| Item 5 | 3 | 4 | 4 | 4 | 2 | 4 | 5 | 0.83 | 0.094 | 0.82 | Excellent |
S-CVI/Ave = (1.00 + 0.83 + 0.67 + 1.00 + 0.83) / 5 = 0.87 (just below .90). S-CVI/UA = 2 of 5 items = 0.40.
Reading it: with six experts, Item 3 (I-CVI .67, k* 0.56) falls below the .78 and Lynn thresholds and would usually be revised or dropped. Items 2 and 5 (I-CVI .83, k* 0.82) meet the thresholds. S-CVI/UA is strict and falls quickly as panels grow, which is why Polit & Beck (2006) suggest reporting S-CVI/Ave and stating which method you used.
Lynn's cut-offs by panel size
Lynn (1986) gave the minimum number of experts who must agree for an item's content validity to exceed chance at the .05 level. The values below are as widely reproduced in the secondary literature (including Polit & Beck 2006); check the original table before citing exact figures.
| Number of experts | Minimum agreeing | Minimum I-CVI |
|---|---|---|
| 3 | 3 of 3 | 1.00 |
| 4 | 4 of 4 | 1.00 |
| 5 | 5 of 5 | 1.00 |
| 6 | 5 of 6 | 0.83 |
| 7 | 6 of 7 | 0.86 |
| 8 | 7 of 8 | 0.88 |
| 9 | 7 of 9 | 0.78 |
| 10 | 8 of 10 | 0.80 |
Lynn recommended a minimum of three experts and suggested that more than ten is probably unnecessary. With five or fewer, every expert must agree.
Good practice and common mistakes
- Say which S-CVI you report. Ave and UA can differ a lot for the same data.
- Do not collapse a 4-point scale after the fact or use an odd-numbered scale with a neutral midpoint; the dichotomy depends on the 4-point design.
- Report the panel: number, expertise and how experts were chosen.
- CVI covers relevance, not clarity or comprehensiveness. Ask experts separately about missing content and wording.
- Round consistently. 5/6 is .833; reporting .83 is fine, but compare against thresholds using the unrounded value.
Free download: CVI worked-example sheet
The formulas, this worked example with every step shown, Lynn's table, the kappa criteria, and a blank rating grid for up to ten experts and fifteen items.
Download the PDF Two A4 pages. No email needed.
Do this in Notion
The free PDF is enough to do the job on paper. If you would rather keep it in one place that updates as you work, the Scale Development Studio is a Notion template that maps items to domains, stores each expert's relevance ratings, and computes I-CVI, modified kappa and S-CVI automatically as ratings come in.
It is a one-time purchase: Core $15, or Plus $24 with the AI pack. See exactly what is inside on the template page, or save by buying it with related templates in a bundle.
Sources
- Lynn MR. Determination and quantification of content validity. Nursing Research 1986;35(6):382-385. pubmed.ncbi.nlm.nih.gov/3640358/
- Polit DF, Beck CT. The content validity index: are you sure you know what's being reported? Research in Nursing & Health 2006;29(5):489-497. doi.org/10.1002/nur.20147
- Polit DF, Beck CT, Owen SV. Is the CVI an acceptable indicator of content validity? Appraisal and recommendations. Research in Nursing & Health 2007;30(4):459-467. pubmed.ncbi.nlm.nih.gov/17654487/
- Cicchetti DV, Sparrow SA. Developing criteria for establishing interrater reliability of specific items. American Journal of Mental Deficiency 1981;86(2):127-137.
Frequently asked questions
How do you calculate the I-CVI?
Count the experts who rate the item 3 or 4 on the 4-point relevance scale and divide by the total number of experts. For example, 5 of 6 experts gives an I-CVI of .83.
What is an acceptable content validity index?
Commonly cited standards are an I-CVI of 1.00 with five or fewer experts, at least .78 with more experts, and an S-CVI/Ave of .90 or higher (Lynn 1986; Polit and Beck 2006).
What is the difference between S-CVI/Ave and S-CVI/UA?
S-CVI/Ave is the mean of the item I-CVIs. S-CVI/UA is the proportion of items that every expert rated 3 or 4. UA is stricter and falls as the panel grows.
What is the modified kappa for content validity?
A chance-adjusted version of the I-CVI proposed by Polit, Beck and Owen (2007): k* = (I-CVI - Pc) / (1 - Pc), where Pc is the binomial probability of chance agreement. Values above .74 are rated excellent.
How many experts do I need for a CVI?
Lynn (1986) recommended at least three experts, and suggested more than ten is probably unnecessary. Six to ten is a common choice.